The question is not only what your probe can measure. It is what stands in front of it.
An excellent sensor on a poor measurement point returns a precise, wrong value. No model, no step up in instrument grade recovers material that is badly presented. That is a bias, and a bias does not average out.
The sampling interface is everything that brings the material in front of the sensor: the nozzle, the window, the flow geometry, the relative movement of probe and product, and the instant of triggering. It is designed before the instrument is chosen, and it holds the part of the result you have the most control over.
Four ways to present the material
None is good in the abstract. Each is good for one geometry, one material and one pace. The in-line and on-line labels are those of the glossary: three of these four interfaces take nothing out of the process.
A probe in front of a window (in-line)
The most common, and the one that calls for the most care. It interrogates a fraction of the stream, and always the same one. It holds where the material is homogeneous at the scale of the question asked, or where the flow mixes enough that the fraction seen keeps changing.
Measurement in free fall (in-line)
At a transfer point the stream presents itself whole and can be traversed edge to edge. It is the configuration most favourable to representativeness where the geometry allows it, and often the cheapest to install, since it calls for no nozzle under pressure. What remains is the density of the falling material: a vertical drop disperses and accelerates it, an inclined plane lets it slide as a steadier bed. A certain angle beats a true vertical.
Scanning the cross-section (in-line)
A probe that moves across the section, or material made to travel past it. Varied positions are then obtained mechanically instead of being hoped for from the flow. It is what a probe on a turning blender already does, though it is rarely described that way.
The by-pass loop (on-line)
The material is brought to the probe rather than the other way round. A caution: the interface then moves the problem without solving it. The tap feeding the loop is what has to be correct, and a by-pass that sorts the material — fines, bubbles, heavy particles — produces a perfectly stable bias. To which is added the transit time through the loop, which has to stay short against the dynamics of the operation.
What makes an interface correct
- The full cross-section, or a variation that covers it. The theory of sampling asks that an increment intercept the entire cross-section of the moving stream. A fixed probe does not; what can stand in its place is a relative movement that ends up exploring the whole section.
- Nothing that sorts the material on the way. A nozzle in a dead zone, a by-pass line that settles, a window that fouls preferentially: each selects a population, and the measurement becomes faithful to that population.
- Triggering tied to the process. Acquiring every thirty seconds makes sense only where the process ignores the clock. A position or phase signal from the equipment serves better, and it is often a minor change to the automation.
- A check that covers the sampling, not the analysis. Repeating the measurement ten times on the same increment says nothing about the batch. Repeating the whole sampling procedure, several times — what the theory of sampling calls the replication experiment — gives the real dispersion, the one no theoretical calculation supplies.
Why this subject always arrives too late
A measurement project almost always starts with the choice of a technology. The interface comes afterwards, once the sensor is bought and the only nozzle available is the one maintenance agreed to drill. The order is inverted: the measurement question sets the pace, the pace sets the installation mode, and the installation mode narrows the technologies that remain. The sensor is chosen last.
That reversal has a visible effect: part of the effort usually spent on instrument performance moves to the design of the measurement point, the mechanical access and the automation. It is also what makes a sampling strategy defensible in front of an auditor, developed on sampling error and on the mass actually analysed.
Describe your measurement point. We will tell you what the probe will really see there.
The nozzle you have in mind, the flow geometry, what moves and what stays still: forty-five minutes is enough to say whether the interface holds, what would have to move, and whether a simpler configuration already exists on your plant. Where the measurement point cannot give a defensible result, you will know before you buy anything.